Last Updated on
Crypto-ML is pleased to announce that all of its trade signals have been now upgraded to version 4 machine learning models. This version is a major enhancement, encompassing a brand-new system for both the predictive capabilities and also the trading mechanisms. This leap forward is the result of years of real-life experience with our machine learning solutions.
Evolution of Bitcoin trading machine learning models
Our previous machine learning models consumed a broad array of technical analysis data, effectively creating a “super-set” technical indicator with parameters that could adapt to changing market conditions via ongoing learning.
The general approach for our previous models is described in this article: How to Use Machine Learning to Trade Bitcoin and Crypto.
As we refined this approach, we added profit capture targets, adaptive stop losses, and trailing stops into the models themselves. This took us from the original release up through the 3.x models.
The weakness of these models, however, is the time spent in catch-up (learning) mode when market conditions changed rapidly.
Despite this warm-up period, the prior models were, of course, no slouches:
That said, we knew we could do better and want to constantly push the envelope.
4.x models: anomaly detection
Based on the experience of developing machine learning models and seeing their results play out in public (Trade History), we were able to develop an entirely new solution that is light years ahead of our original approach. It is an evolutionary leap.
Fundamentally, the new models monitor a large set of data, looking for anomalies that typically precede unusually large movements in the market. This approach works in nearly any market condition.
The full detail of how the 4.x models work is described here: Bitcoin Trading with Machine Learning Anomaly Detection.
This approach was introduced back in October, 2019 as the Bitcoin Beta Model. It successfully exited beta, as described in Bitcoin Machine Learning Model Upgraded to 4.1.
Additional upgrades: ETH, BCH, and LTC
As of January 31, 2020, the other long machine learning models have also been upgraded to the 4.x architecture, including:
- Bitcoin Cash
Each incorporates the learnings from the BTC Beta period.
What to expect from the upgrades
Just as with the Bitcoin upgrade, the Etheruem, Bitcoin Cash, and Litecoin models will also experience an evolutionary leap in capabilities.
The statistical measures and test results are exceptional. While each of these new models produce excellent results, we can put the performance expectations in tiers:
- Bitcoin and Litecoin
- Ethereum and Bitcoin Cash
That is, the models behind Ethereum and Bitcoin Cash demonstrate even better predictive capabilities than our BTC Beta model. We expect this to translate to better trades.
Regardless, all are great performers and we need to see the results unfold in real life. Additionally, overall market direction matters. It’s still a best practice to trade the overall best-performing asset.
Stop losses updated
The stop losses are now set to match the trading optimizations of the new models.
Please note that these values will not change as frequently as before. Since the new models work in such a broad range of market conditions, you’ll see less adaptation required.
What do you need to do?
Nothing! If you’re on our Trader or Auto Trade membership, you’ll notice a few differences in our Member Dashboard. But this was an in-place upgrade, meaning the backend has changed, but the customer-facing portions appear largely the same.
But if you’re happy with the results, you can always share with a friend using our referral program.
What’s next? You help determine!
In terms of housekeeping, there are other updates coming related to this upgrade:
- Revised Member Dashboard
- Updated Visualizations
- Revision to text and guides throughout the site
But on the more exciting side, we have the question of what our machine learning team should focus on next.
You’ll notice Bitcoin Short is not in the list of items upgraded to 4.x. The Short system is temporarily paused. We will bring the Short signal back, but want to understand the priorities of our customers first.
There are several projects our machine learning team can take on next. We plan on putting the priority to a customer vote over the next week.
So watch for the upcoming customer poll and let your voice be heard!